Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Denoising via Repainting: an image denoising method using layer wise medical image repainting

Arghya Pal, Sailaja Rajanala, CheeMing Ting, Raphael Phan

TL;DR

This paper tackles MRI denoising by addressing the trade-off between noise suppression and preservation of fine anatomical detail. It introduces a multi-scale framework that combines an anisotropic Gaussian scale-space with progressive Bezier-path repainting, enabling coarse-to-fine reconstruction of image components. Key contributions include a data-efficient, cross-domain denoising pipeline that does not rely on extensive labeled data and utilizes MSE and Xing-loss to optimize Bezier path representations across scales. Experiments across AxFLAIR, Cor-PD knee, FastMRI, and Modl datasets show PSNR/SSIM gains and robust structural preservation, with future work aiming to extend to 3D data and incorporate semantic guidance for improved reconstructions.

Abstract

Medical image denoising is essential for improving the reliability of clinical diagnosis and guiding subsequent image-based tasks. In this paper, we propose a multi-scale approach that integrates anisotropic Gaussian filtering with progressive Bezier-path redrawing. Our method constructs a scale-space pyramid to mitigate noise while preserving critical structural details. Starting at the coarsest scale, we segment partially denoised images into coherent components and redraw each using a parametric Bezier path with representative color. Through iterative refinements at finer scales, small and intricate structures are accurately reconstructed, while large homogeneous regions remain robustly smoothed. We employ both mean square error and self-intersection constraints to maintain shape coherence during path optimization. Empirical results on multiple MRI datasets demonstrate consistent improvements in PSNR and SSIM over competing methods. This coarse-to-fine framework offers a robust, data-efficient solution for cross-domain denoising, reinforcing its potential clinical utility and versatility. Future work extends this technique to three-dimensional data.

Denoising via Repainting: an image denoising method using layer wise medical image repainting

TL;DR

This paper tackles MRI denoising by addressing the trade-off between noise suppression and preservation of fine anatomical detail. It introduces a multi-scale framework that combines an anisotropic Gaussian scale-space with progressive Bezier-path repainting, enabling coarse-to-fine reconstruction of image components. Key contributions include a data-efficient, cross-domain denoising pipeline that does not rely on extensive labeled data and utilizes MSE and Xing-loss to optimize Bezier path representations across scales. Experiments across AxFLAIR, Cor-PD knee, FastMRI, and Modl datasets show PSNR/SSIM gains and robust structural preservation, with future work aiming to extend to 3D data and incorporate semantic guidance for improved reconstructions.

Abstract

Medical image denoising is essential for improving the reliability of clinical diagnosis and guiding subsequent image-based tasks. In this paper, we propose a multi-scale approach that integrates anisotropic Gaussian filtering with progressive Bezier-path redrawing. Our method constructs a scale-space pyramid to mitigate noise while preserving critical structural details. Starting at the coarsest scale, we segment partially denoised images into coherent components and redraw each using a parametric Bezier path with representative color. Through iterative refinements at finer scales, small and intricate structures are accurately reconstructed, while large homogeneous regions remain robustly smoothed. We employ both mean square error and self-intersection constraints to maintain shape coherence during path optimization. Empirical results on multiple MRI datasets demonstrate consistent improvements in PSNR and SSIM over competing methods. This coarse-to-fine framework offers a robust, data-efficient solution for cross-domain denoising, reinforcing its potential clinical utility and versatility. Future work extends this technique to three-dimensional data.

Paper Structure

This paper contains 11 sections, 4 equations, 4 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Step 1: Scale-Space Pyramid
  4. Anisotropic Gaussian Blur.
  5. Scale-Space Construction.
  6. Step 2: Progressive Component Redrawing in $I_{0}$
  7. Bezier-Path Representation.
  8. Step 3: Iteration & Optimization Across Scales
  9. Step 4: Loss Functions
  10. Results
  11. Discussion and Conclusion

Figures (4)

  • Figure 1: (Top) Our methodology (§\ref{['sec_method']}) progressively adds Bezier paths for each component in a coarse-to-fine manner. (Bottom) Demonstration of denoising an image in a layer-wise coarse-to-fine manner with redrawing using a small number of paths. $c_1,\cdots,c_9$ indicate the path numbers.
  • Figure 2: Anisotropic Gaussian Blur with varying $G(\mu_t, \sigma_t)$perona1990scale. Larger values of $\mu_t$ and $\sigma_t$ yield stronger smoothing, while smaller values preserve finer details.
  • Figure 3: (Top) AxFLAIR brain MRI, (Bottom) Cor-PD knee MRI; qualitative comparison of DnCNN 7839189, N2N pmlr-v80-lehtinen18a, Ne2Ne Huang_2021_CVPR, Restoformer zamir2022restormer, N2void krull2019noise2void, LAN 10656535, and our method. First column represent ground truth image.
  • Figure 4: (Top) FastMRI knee reconstruction, (Bottom) Modl aggarwal2021model Brain MRI; qualitative comparison of Ne2Ne Huang_2021_CVPR, Restoformer zamir2022restormer, N2void krull2019noise2void, LAN 10656535, and our method. First column represent ground truth image.